beam-ai-team/beam-next-skills-audit

Beam Next Skills Registry Audit

You are auditing this beam-next-skills repository: the canonical index is registry.yaml at the repo root; each skill is a folder under skills/ containing at least SKILL.md with YAML frontmatter.

Goal: Find drift (registry vs filesystem), bad or missing metadata, duplicates, and content quality issues that hurt discovery in Beam Next / Prism.

---

Architecture (what you are auditing)

beam-next-skills/
├── registry.yaml          # Machine-readable index (skill_count, name, path, version, …)
├── README.md              # Human overview (counts can drift from registry)
├── scripts/
│   └── validate_skill_frontmatter.py   # Phase-1 frontmatter rules
└── skills/
    ├── {category}/
    │   └── {skill-name}/
    │       └── SKILL.md   # Source of truth for description triggers; frontmatter required

Beam Next: A local workspace may mirror these under 00-system/skills/; this audit is for this Git repo unless another root is given.

---

Step 0: Resolve repository root

• Default: the root of the beam-next-skills clone (directory that contains registry.yaml and skills/).
• Set SKILLS_ROOT to that path for all commands below.

Verify:

test -f "$SKILLS_ROOT/registry.yaml" && test -d "$SKILLS_ROOT/skills" && echo OK

---

Step 1: Gather audit scope

| Scope | What you run | |-------|----------------| | Full | Everything below (default) | | Registry + disk | Steps 2–4 only (drift, orphans, phantoms) | | Frontmatter | Step 3 script + Step 5 | | Quality / consolidation | Steps 6–7 |

If unspecified, run full.

---

Step 2: Registry internal consistency

From $SKILLS_ROOT:

1. skill_count vs actual entries

   grep -E "^skill_count:" registry.yaml
   grep -c "^- name:" registry.yaml
   

   The count of - name: lines under skills: must match skill_count. If not, registry is stale.

2. Duplicate skill names

   grep "^- name:" registry.yaml | sed 's/.*name: //' | sort | uniq -d
   

   Any output = duplicate names (must fix).

3. Duplicate paths

   grep "path: skills/" registry.yaml | sort | uniq -d
   

4. Missing path or broken paths

   For each path: in registry.yaml, verify the folder exists and contains SKILL.md:

   while read -r p; do
     test -f "$SKILLS_ROOT/$p/SKILL.md" || echo "MISSING: $p"
   done < <(grep "path: skills/" registry.yaml | sed 's/.*path: //')
   

---

Step 3: Automated frontmatter validation

cd "$SKILLS_ROOT" && python3 scripts/validate_skill_frontmatter.py

CI-style failure:

python3 scripts/validate_skill_frontmatter.py --strict

Rules enforced (summary):

• SKILL.md must start with --- frontmatter.
• Prefer type: skill (or agent per project conventions).
• tags: 2–5 items.
• category must be one of the allowed set in the script.

---

Step 4: Disk vs registry (orphans and phantoms)

1. Phantom registry entries — path in registry.yaml but no SKILL.md on disk (Step 2.4).

2. Orphan skills — folder with SKILL.md under skills/ but no matching path in registry.yaml.

   find "$SKILLS_ROOT/skills" -name SKILL.md -print | sed "s|$SKILLS_ROOT/||; s|/SKILL.md||" | sort > /tmp/disk-skills.txt
   # Diff against paths extracted from registry.yaml
   

   Report: orphans (on disk only) and phantoms (registry only).

---

Step 5: Metadata drift (registry vs SKILL.md)

For a sample of skills (or all on a full audit), compare:

• name in registry vs name in frontmatter — must match.
• version / updated — flag large drift.
• visibility — must be public or team in SKILL.md and should match registry.yaml (team = internal-only, e.g. ROI/costs; public = catalog-safe).

Skills missing type: in frontmatter:

grep -L "^type:" $(find "$SKILLS_ROOT/skills" -name SKILL.md) 2>/dev/null | head

Align visibility from registry into SKILL.md (optional bulk fix):

python3 scripts/sync_visibility_from_registry.py --dry-run
python3 scripts/sync_visibility_from_registry.py

---

Step 6: Description & trigger quality (spot audit)

For each category or a random sample of 15–20 skills, open SKILL.md and check:

| Check | Pass | |-------|------| | Trigger clarity | description in frontmatter says when to load (phrases, user intent). | | Not a stub | Body is not only a title; has workflow or clear instructions. | | Connector skills | *-connect skills explain setup + link to *-master or product skills when relevant. | | Consolidated skills | Large integrations describe CLI/resources instead of duplicating many endpoints. |

---

Step 7: Structural anti-patterns

• Script-only folders — scripts under skills/ but no SKILL.md.
• Empty or stub SKILL.md — placeholder only.
• Broken relative paths in docs — old folder names vs current skills/… layout.

---

Step 8: Generate audit report

# Beam Next Skills Registry Audit — {date}

## Summary
- Registry path: `{SKILLS_ROOT}`
- `skill_count` in YAML: {n}
- `- name:` entries: {n}  → MATCH / MISMATCH
- Duplicate names: {n}
- validate_skill_frontmatter.py issues: {n}
- Orphan skills (disk, not registry): {n}
- Phantom entries (registry, no SKILL.md): {n}

## Critical issues
{Duplicates, phantoms, skill_count mismatch}

## Warnings
{Frontmatter warnings, missing type, tag/category count}

## Description quality (sample)
| Skill | Trigger clarity | Notes |
|-------|-----------------|-------|
| ... | Good / Needs work / Bad | ... |

## Structural notes
{Script-only, outdated paths, consolidation candidates}

## Recommended actions
1. {Highest priority}
2. ...

---

Verification

1. Duplicate check: grep "^- name:" registry.yaml | sed 's/.*name: //' | sort | uniq -d — must be empty.
2. If registry was edited: skill_count matches entry count.

Optional:

cd "$SKILLS_ROOT" && python3 build_registry.py .   # if build_registry.py exists

---

Guidelines

• Read real files — README skill counts can drift from registry.yaml.
• Actionable only — prioritize phantoms, duplicates, and skill_count mismatch.
• Align with scripts/validate_skill_frontmatter.py — extend that script if you add new automated rules.

---

Gotchas

1. skill_count is manual — easy to forget after adding a skill.
2. Nested skill folders — resolve orphans using full path: strings, not only top-level names.
3. Registry description may be shorter than SKILL.md — OK unless triggers are wrong.
4. validate_skill_frontmatter.py does not validate description length or trigger phrases — spot-audit only (Step 6).